If you're just pointing the meter and the camera randomly at a scene they would be pretty unlikely to agree. The camera is probably using some sort of matrix algorithm and is basing it's reading according to the needs of the sensor, not a piece of film.
If you meter from something like an evenly lit wall from a few inches, making sure you don't have a shadow on it, they may be a little closer. If you have a grey card, that would be the ideal target. That the meter and the camera agree when in incident mode is an indication that your comparison technique for the reflective reading may not be valid.